Historical carbon dioxide emissions caused by land-use changes are possibly larger than assumed

HAL Id: hal-01681571

https://hal.archives-ouvertes.fr/hal-01681571

HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Historical carbon dioxide emissions caused by land-use

changes are possibly larger than assumed

Almut Arneth, Stephen Sitch, Julia Pongratz, B. D. Stocker, Philippe Ciais,

B. Poulter, A. D. Bayer, Alberte Bondeau, L. Calle, L. P. Chini, et al.

To cite this version:

Almut Arneth, Stephen Sitch, Julia Pongratz, B. D. Stocker, Philippe Ciais, et al.. Historical carbon dioxide emissions caused by land-use changes are possibly larger than assumed. Nature Geoscience, Nature Publishing Group, 2017, 10 (2), pp.79-84. �10.1038/NGEO2882�. �hal-01681571�

2

(10) International Centre for Water Resources and Global Change, hosted by the German Federal 23

Institute of Hydrology. Am Mainzer Tor 1, 56068 Koblenz, Germany 24

(11) College of Engineering, Mathematics and Physical Sciences, University of Exeter, Exeter, EX4 25

4QE, UK 26

(12) The Institute of Applied Energy, Minato, Tokyo 105-0003, Japan 27

(13) Dept of Physical Geography and Ecosystem Science, Sölvegatan 12, Lund University,

28

22362 Lund, Sweden

29

(14) School of Geography, Earth & Environmental Sciences and Birmingham Institute of Forest 30

Research, University of Birmingham, Birmingham, B15 2TT, United Kingdom 31

(15) Met Office Hadley Centre, FitzRoy Road, Exeter, EX1 3PB, UK 32

(16) Max Planck Institute for Biogeochemistry, 07701 Jena, Germany 33

34

35

3

The terrestrial biosphere absorbs about 20% of fossil fuel CO2 emissions. The overall

37

magnitude of this sink is constrained by the difference between emissions, the rate of 38

increase in atmospheric CO2 concentrations and the ocean sink. However, the land sink

39

is actually composed of two largely counteracting fluxes that are poorly quantified: 40

fluxes from land-use change and CO2 uptake by terrestrial ecosystems. Dynamic global

41

vegetation model simulations suggest that CO2 emissions from land-use change have

42

been substantially underestimated because processes such as tree harvesting and land-43

clearing from shifting cultivation have not been considered. Since the overall terrestrial 44

sink is constrained, a larger net flux as a result of land-use change implies that 45

terrestrial uptake of CO2 is also larger, and that terrestrial ecosystems might have

46

greater potential to sequester carbon in the future. Consequently, reforestation projects 47

and efforts to avoid further deforestation could represent important mitigation 48

pathways, with co-benefits for biodiversity. It is unclear whether a larger land carbon 49

sink can be reconciled with our current understanding of terrestrial carbon cycling. In 50

light of our possible underestimation of the historical residual terrestrial carbon sink 51

and associated uncertainties, we argue that projections of future terrestrial carbon 52

uptake and losses are more uncertain than ever. 53

54

The net atmosphere-to-land carbon flux (FL) is typically inferred as the difference between relatively

55

well-constrained terms of the global carbon cycle: fossil fuel and cement emissions, oceanic carbon 56

uptake and atmospheric growth rate of CO2 (see Textbox) 1

. In contrast, very large uncertainties exist 57

in how much anthropogenic land-use and land-cover change (FLULCC) contributes to FL, which

58

propagates into large uncertainties in the estimation of the residual carbon flux (FRL) (Box 1). The lack

59

of confidence in separating FL into its component fluxes diminishes the predictive capacity for

60

terrestrial carbon cycle projections into the future. It restricts our ability to estimate the capacity of 61

4

land ecosystems to continue to mitigate climate change, and to assess land management options for 62

land-based mitigation policies. 63

As land-use change emissions and the residual sink are spatially closely enmeshed, global-scale 64

observational constraints do not exist for estimating FLULCC or FRL separately. Dynamic Global

65

Vegetation Models (DGVMs) have over recent years been used to infer the magnitude and spatial 66

distribution of FLULCC as well as of FRL, while FLULCC has traditionally been also derived from

data-67

driven approaches such as the bookkeeping method 1-3 (Box 1). Although large, for some sources of 68

uncertainties in FLULCC (such as differences in baseline years used for calculation, how environmental

69

effects have been considered, or assumptions about wood products) there is no good reason to believe 70

that these would introduce a systematic under- or overestimation4-6. However, until recently, most 71

processes related to land management and the subgrid-scale dynamics of land-use change have been 72

ignored in large-scale assessments of the terrestrial carbon balance, and we argue here that including 73

these missing processes might systematically increase the magnitude of FLULCC. In turn, an upward

74

revision of FLULCC implies through the global budget the existence of a substantially higher FRL and

75

raises the question whether this is plausible given our understanding of the response of ecosystems to 76

changing environmental conditions. 77

Gross land-cover transitions such as shifting cultivation 78

Opposing changes in different land-use types can take place simultaneously within a region (see 79

Methods and Supplementary Figure 1), for example: an area might be converted from natural to 80

managed land, whereas an equal area within the same region might be abandoned or reforested, 81

equating to a net-zero land-cover change. The magnitude of these bi-directional changes depends on 82

the size of the area investigated. Over thousands of kilometres squared (the typical resolution of 83

DGVMs), ignoring sub-grid changes can have a substantial effect on the simulated carbon cycle, since 84

accounting for the gross changes (for example, th6e parallel conversion to, and abandonment of, 85

agricultural land in the same grid-cell) includes (rapid) carbon losses from deforestation, (slow) loss 86

from post-deforestation soil-legacy effects, and (slow) uptake in areas of regrowth. In sum this leads to 87

5

younger mean stand age, smaller biomass pools and thus higher FLULCC compared to net area-change

88

simulations. 89

Gross area transitions are fundamental to land-use land-cover change (LULCC) dynamics in areas of 90

shifting cultivation (SC) in the tropics7, but also occur elsewhere8. Gross forest loss far exceeding net 91

area loss can be demonstrated from remote-sensing products globally9, although these products in 92

themselves cannot distinguish effects of logging from natural disturbance events such as fire or 93

storms. Secondary forests in the tropics can return to biomass carbon stocks comparable to old-growth 94

forest within five to six decades10, but the same is not the case for soil carbon. Also, fallow lengths in 95

SC systems tends to be shorter, and show a decreasing trend in many regions11. These dynamics result 96

in the degraded vegetation and reduced soil carbon stocks commonly observed in disturbed forest land 97 12 . 98 Wood harvesting 99

Until recently, global DGVM studies that accounted for LULCC concentrated on the representation of 100

conversion of natural lands to croplands and pastures, while areas under forest cover were represented 101

as natural forest, and hence by each model’s dynamics of establishment, growth and mortality. Two-102

thirds to three-quarters of global forests have been affected by human use, which is mainly due to 103

timber harvest; but forests are also a source of firewood or secondary products; or used for recreational 104

purposes13. Between 1700-2000 an estimated 86 PgC has been removed globally from forests due to 105

wood harvesting (WH)14 . WH leads to reduced carbon density on average in managed forests15 and 106

can ultimately result in degradation in the absence of sustainable management strategies. Furthermore, 107

the harvest of wood can reduce litter input, which lowers soil pools13. Bringing a natural forest under 108

any harvesting regime probably will lead to net-CO2 emissions to the atmosphere – with a magnitude

109

and time-dependency conditionnal on harvest intensity and frequency, regrowth and the fate and 110

residence time of the wood products. 111

6

Grazing and crop harvesting, and cropland management 112

Management is not only fundamental for the carbon balance of forests, but also for pasture and 113

cropland. As with forests, accounting for management processes on arable lands has only recently 114

been included in DGVMs (see Methods). Regular grazing and harvesting (GH), and more realistic 115

crop management processes (MC) such as flexible sowing and harvesting, or tillage, will enhance 116

FLULCC (ref. 16). Over decadal timescales, conversion of forest to cropland has been observed to reduce

117

soil carbon pools by around 40% (ref. 17), resulting from reduced vegetation litter soil inputs and 118

enhanced soil respiration in response to tillage, although the effect and magnitude of the latter is being 119

debated18 . Conversion to pasture often has either little effect, or may even increase soil carbon17. 120

Impacts of land-management processes on the carbon cycle 121

The few published DGVM studies that account for the management of land more realistically16,19-21 122

consistently suggest a systematically larger FLULCC over the historical period compared to estimates

123

that ignored these processes, with important implications for our understanding of the terrestrial 124

carbon cycle and its role for historical (and future) climate change. In order to assess if results from 125

these initial experiments hold despite differences among models, we compile here results from a wider 126

set of DGVMs (and one DGVM “emulator”, see Methods and Supplementary Table 1), adopting the 127

approach described in ref. 2. FLULCC was calculated as the difference between a simulation in which

128

CO2 and climate were varied over the historical period, at constant (pre-industrial) land use, and one in

129

which land use was varied as well. 130

When accounting for SC and WH, FLULCC was systematically enhanced (Fig. 1).

Shifting-131

cultivation, assuming that no shade-trees remain in cultivated areas, results in increased cumulative 132

FLULCC over the period 1901-2014 on average by 35 ± 18 PgC (Fig. 1, Supplementary Table 2).

133

Although three DGVMs had demonstrated this effect previously19-21, an upward shift of FLULCC was

134

also found in the other models that performed additional SC simulations for this study. Including WH 135

caused FLULCC to increase over the same time period by a similar magnitude to SC: 30 ± 21 PgC.

7

Trends in WH-related FLULCC over time differed between models (Fig. 1) probably due to different

137

rates of post-harvest regrowth, and assumptions about residence time in different pools22. Including the 138

harvesting of crops and the grazing of pastures also resulted in larger FLULCC, as carbon harvested or

139

grazed is consumed and released as CO2 rapidly instead of decaying slowly as litter and soil organic

140

matter. Beyond harvest, accounting for more realistic MC such as tillage processes also showed, with 141

one exception (in which tillage effects were not modelled, see methods) an enhancement of FLULCC

142

emissions. 143

When ignoring the additional land-use processes investigated here, average FLULCC is 119 ± 50 PgC

144

(Supplementary Table 2). Adding effects of SC, WH, GH and MC enhance land-use change emissions 145

by, on average, 20-30% each (Fig. 2; Supplementary Table 1), with individually large uncertainties. 146

The combined effects on FLULCC are difficult to judge as models do not yet account for all land-use

147

dynamics. For instance, SC and WH effects are expected to enhance FLULCC additively as there is little

148

overlap in the input dataset used by DGVMs regarding the areas that are assumed to be under SC, and 149

areas where other types of forest harvesting occur7. But in the case of accounting for harvesting and 150

other management on arable lands and pastures, carbon cycle interactions with SC and WH cannot be 151

excluded because subsequent transitions could occur in a grid location, between primary vegetation 152

and cropland, pastures or secondary forests. The overall enhancement of FLULCC therefore, will need to

153

be explored with model frameworks that include all dynamic land-use-change processes. DGVMs 154

currently contributing to the annual update of the global carbon budget account for some of the 155

processes examined here, but as of yet not at all comprehensively, and we thus expect DGVM-based 156

FLULCC to increase substantially compared to results reported in ref. 1. As a consequence, the

157

discrepancy to bookkeeping estimates of FLULCC will become larger, although results in ref. 23 call for

158

a broader range of bookkeeping approaches as well. 159

Implications for the historical residual land sink 160

In order to match FL in the global carbon budget (Box 1) for the historical period a substantially larger

161

FLULCC would need to be balanced by a corresponding increase in FRL, which could be either due to

8

underestimated historical increase in gross primary production (GPP) and vegetation biomass, 163

overestimated heterotrophic carbon loss, or both. The question arises if such a discrepancy is credible 164

in light of today’s understanding. For instance, by compiling a number of observations Pan et al.24 165

suggested a forest sink that is in line with total carbon budget estimates1. However, their study 166

excluded savannahs, grasslands, and woodlands and in semi-arid regions alone carbon uptake was 167

estimated to be about 20% of the terrestrial sink (plus around another 30% from other non-forested 168

ecosystems), which also dominate the recent positive trend in carbon uptake25. Reconstructing the 169

Austrian historical forest sink from inventory data also suggested a much larger residual sink, 170

compared with (bookkeeping) model results26. 171

The response of photosynthesis to increasing CO2 could underlie more than half of today’s land carbon

172

sink27. Several recent lines of observation-based evidence suggest that GPP may have undergone much 173

stronger enhancement over the last century than currently calculated by DGVMs. These studies 174

include isotopic analysis of herbarium plant samples, of stable oxygen isotope ratios in atmospheric 175

CO2, and accounting for the effect of leaf mesophyll resistance to CO2 (refs 28-30). Ciais et al. 31

176

inferred a pre-industrial GPP of 80 PgC a-1 based on measurements of oxygen isotopes in ice-core air, 177

indicative for a 33% difference to the often-used present-day GPP benchmark of ~120 PgC a-1 (ref. 32) 178

and independently consistent with the 35% increase suggested by ref. 28. In contrast, the participating 179

DGVMs in this study show an average increase of GPP by only 15% between the first and last ten 180

years of the simulation (not shown). 181

Whether or not enhancements in GPP translate into increased carbon storage depends on other factors 182

such as nutrient and water supply, seen for instance in the mixed trends in stem growth found in forest 183

inventories33,34. Much work remains to better understand the response of ecosystem carbon storage to 184

increasing atmospheric CO2 concentration 35

. Ultimately, carbon turnover time determines whether or 185

not enhanced growth will only result in increasing carbon pools22. Besides GPP and heterotrophic 186

ecosystem respiration, lateral carbon flows play an important role in the ecosystem carbon sink. 187

Recent syntheses that combined a range of observations, inventories of carbon stock changes, trade 188

9

flows and transport in waterways, estimated dissolved organic carbon losses to account for a flux of > 189

1.0 PgC a-1, with an unknown historical trend36,37. The fate of this carbon is highly uncertain, but its 190

inclusion would enhance the calculated residual sink via an additional loss term (Box 1, equation (1)). 191

Taken together, a number of candidates for underestimated FRL in today’s models are plausible, and a

192

combination of the above listed processes likely. It remains to be seen whether a larger FLULCC can be

193

supported by observation-based estimates. Several lines of evidence suggest that a common low-bias 194

in the historic FLULCC could affect all DGVMs, and the challenge of resolving the many open issues

195

will stay with us for some years to come. 196

Unknowns in historical LULCC reconstructions 197

Patterns and historical trends of deforestation, cropland and pasture management, or WH are 198

uncertain. Land-use reconstructions differ substantially in terms of the time, location and rate of 199

LULCC (see ref. 38 and reference therein). The DGVM and climate science community has mostly 200

relied on the LUH1 dataset by Hurtt et al.7, chiefly because it provides the needed seamless time series 201

from the historical period into future projections at the spatial resolution required by DGVMs. Clearly 202

such a globally applicable gridded dataset must necessarily include simplifications. For instance, the 203

assumed uniform 15-year turnover in tropical SC systems7 cannot account for the known variation 204

between a few years and one to two decades, or trends towards shorter fallow periods in some regions 205

(see ref. 11 and references therein), although there is also an increasing proportion of permanent 206

agriculture. Likewise, not only the amount of WH but also the type of forestry (coppice, clear-cut, 207

selective logging, fuel-wood) will vary greatly in time and space, which is difficult to hindcast39,40. 208

In upcoming revisions to LUH1 (LUH-2, http://luh.umd.edu/data.shtml), forest-cover gross 209

transitions are now constrained by the remote sensing information9, and have overall been re-estimated 210

(Fig. 3). Whether or not this will result in reduced SC carbon loss estimates in recent decades remains 211

to be seen. At the same time, these historical estimates consider large gross transitions of land-cover 212

change only for tropical regions even though there is good reason to believe that bi-directional 213

changes occur elsewhere41. For Europe alone, a recent assessment that is relatively impartial to spatial 214

10

resolution estimated twice the area having undergone land-use transitions since 1900 when accounting 215

for gross vs. net area changes8. This leads to substantial increase in the calculated historical European 216

FLULCC, both in a bookkeeping-model and DGVM-based study 42

. Historical land carbon cycle 217

estimates therefore are not only highly uncertain due to missing LULCC processes, but equally so due 218

to the LULCC reconstructions per se. However, for a given reconstruction, accounting for additional 219

processes discussed here will always introduce a unidirectional enhancement in FLULCC compared to

220

ignoring these processes. 221

Implications for the future land carbon mitigation potential 222

Our calculated increases in FLULCC, in absence of a clear understanding of the processes underlying

223

FRL, notably strengthen the existing arguments to avoid further deforestation (and all ecosystem

224

degradation) – an important aspect of climate change mitigation, with considerable co-benefits to 225

biodiversity and a broad range of ecosystem service supply. One could also conjecture whether or not 226

a larger historical carbon loss through LULCC would imply a larger potential to sequester carbon 227

through reforestation, than thought so far. However, assessments of mitigation potentials must 228

consider the often relatively slow carbon gain in regrowing forests (compared to the rapid, large loss 229

during deforestation), in particular the sluggish replenishment of long-term soil carbon storage43,44. 230

What is more, trees grow now, and will in future, under very different environmental conditions 231

compared to the past. A warmer climate increases mineralisation rates and hence enhances nutrient 232

supply to plant growth, supporting the CO2 fertilisation effect, but also stimulates heterotrophic decay

233

of existing soil carbon and/or flow of dissolved carbon, with as yet no agreement about the net 234

effects3,45. Regrowing forests might also in future be more prone to fire risk, and other episodic events 235

such as wind-throw or insect outbreaks46,47, crucial ecosystem features not yet represented well in 236

models48. This question of “permanence” has been an important point of discussion at conferences 237

under the UNFCCC, and also endangers the success of payment-for-ecosystem-services schemes that 238

target conservation measures, as it is unclear how an increasing risk of losing carbon-uptake potential 239

can be accounted for49,50. 240

11

Given that we may be greatly underestimating the present-day FRL, and therefore missing or

241

underestimating the importance of key driving mechanisms, projections of future terrestrial carbon 242

uptake and losses appear more fraught with uncertainty than ever. In the light of the findings 243

summarized here, this poses not only a major challenge when judging mitigation efforts, but also for 244

the next generation of DGVMs and Earth System models to assess the future global carbon budget. 245

Future work therefore needs to concentrate on representing the interactions between physiological 246

responses to environmental change in ecosystems with improved representations of human land 247 management. 248 249 References 250 251

1 Le Quere, C. et al. Global Carbon Budget 2015. Earth Sys. Sci. Data 7, 349-396 (2015). 252

2 Sitch, S. et al. Recent trends and drivers of regional sources and sinks of carbon dioxide. 253

Biogeosciences 12, 653-679 (2015). 254

3 Ciais, P. et al. in Climate Change 2013: The Physical Science Basis. Contribution of Working 255

Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change.

256

(eds T.F. Stocker et al.) (Cambridge University Press, 2013). 257

4 Pongratz, J., Reick, C., Houghton, R. A. & House, J. I. Terminology as a key uncertainty in 258

net land use flux estimates. Earth Syst. Dyn. 5, 177-195 (2013). 259

5 Gasser, T. & Ciais, P. A theoretical framework for the net land-to-atmosphere CO2 flux and its

260

implications in the definitions of "emissions from land-use change". Earth Syst. Dyn. 4, 171-261

186 (2013). 262

6 Houghton, R. A. et al. Carbon emissions from land use and land-cover change. 263

Biogeosciences 9, 5125-5142 (2012). 264

12

7 Hurtt, G. C. et al. Harmonization of land-use scenarios for the period 1500-2100: 600 years of 265

global gridded annual land-use transitions, wood harvest, and resulting secondary lands. Clim. 266

Change 109, 117-161 (2011). 267

8 Fuchs, R., Herold, M., Verburg, P. H., Clevers, J. G. P. W. & Eberle, J. Gross changes in 268

reconstructions of historic land cover/use for Europe between 1900 and 2010. Glob. Change 269

Biol. 21, 299-313 (2015). 270

9 Hansen, M. C., Stehman, S. V. & Potapov, P. V. Quantification of global gross forest cover 271

loss. Proc. Nat. Acad. Sci. 107, 8650-8655 (2010). 272

10 Poorter, L. et al. Biomass resilience of Neotropical secondary forests. Nature 530, 211-214 273

(2016). 274

11 van Vliet, N. et al. Trends, drivers and impacts of changes in swidden cultivation in tropical 275

forest-agriculture frontiers: A global assessment. Glob. Env. Change 22, 418-429 (2012). 276

12 Grace, J., Mitchard, E. & Gloor, E. Perturbations in the carbon budget of the tropics. Glob. 277

Change Biol. 20, 3238-3255 (2014). 278

13 Erb, K.-H. et al. Land management: data availability and process understanding for global 279

change studies. Glob. Change Biol., gcb.13443 (2016). 280

14 Hurtt, G. C. et al. The underpinnings of land-use history: three centuries of global gridded 281

land-use transitions, wood-harvest activity, and resulting secondary lands. Glob. Change Biol., 282

12, 1208-1229 (2006). 283

15 Noormets, A. et al. Effects of forest management on productivity and carbon sequestration: A 284

review and hypothesis. For. Ecol. Manag. 355, 124-140 (2015). 285

16 Pugh, T. A. M. et al. Carbon emission from land-use change is substantially enhanced by 286

agricultural management. Env. Res. Lett., 124008 (2015). 287

13

17 Guo, L. B. & Gifford, R. M. Soil carbon stocks and land use change: a meta analysis. Glob. 288

Change Biol. 8, 345-360 (2002). 289

18 Powlson, D. S. et al. Limited potential of no-till agriculture for climate change mitigation. 290

Nat. Clim. Change 4, 678-683 (2014). 291

19 Shevliakova, E. et al. Carbon cycling under 300 years of land use change: Importance of the 292

secondary vegetation sink. Glob. Biogeochem. Cycles 23, GB2022 (2009). 293

20 Stocker, B. D., Feissli, F., Strassmann, K. M., Spahni, R. & Joos, F. Past and future carbon 294

fluxes from land use change, shifting cultivation and wood harvest. Tellus B 66, 23188 (2014). 295

21 Wilkenskjeld, S., Kloster, S., Pongratz, J., Raddatz, T. & Reick, C. H. Comparing the 296

influence of net and gross anthropogenic land-use and land-cover changes on the carbon cycle 297

in the MPI-ESM. Biogeosciences 11, 4817-4828 (2014). 298

22 Friend, A. D. et al. Carbon residence time dominates uncertainty in terrestrial vegetation 299

responses to future climate and atmospheric CO2. Proc. Nat. Acad. Sci. 111, 3280-3285 300

(2014). 301

23 Hansis, E., Davis, S. J. & Pongratz, J. Relevance of methodological choices for accounting of 302

land use change carbon fluxes. Glob. Biogeochem. Cycles 29, 1230-1246 (2015). 303

24 Pan, Y. et al. A Large and Persistent Carbon Sink in the World's Forests. Science 333, 988-304

993 (2011). 305

25 Ahlström, A. et al. The dominant role of semi-arid ecosystems in the trend and variability of 306

the land CO2 sink. Science 348, 895-899 (2015). 307

26 Erb, K.-H. et al. Bias in attributing of forest carbon sinks. Nat. Clim. Change 3, 854-856 308

(2013). 309

14

27 Schimel, D., Stephens, B. B. & Fisher, J. B. Effect of increasing CO2 on the terrestrial carbon 310

cycle. Proc. Nat. Acad. Sci. 112, 436-441 (2015). 311

28 Ehlers, I. et al. Detecting long-term metabolic shifts using isotopomers: CO2-driven 312

suppression of photorespiration in C-3 plants over the 20th century. Proc. Nat. Acad. Sci. 112, 313

15585-15590 (2015). 314

29 Sun, Y. et al. Impact of mesophyll diffusion on estimated global land CO2 fertilization. 315

Proceedings of the National Academy of Sciences, doi:10.1073/pnas.1418075111 (2014). 316

30 Welp, L. R. et al. Interannual variability in the oxygen isotopes of atmospheric CO2 driven by 317

El Nino. Nature 477, 579-582 (2011). 318

31 Ciais, P. et al. Large inert carbon pool in the terrestrial biosphere during the Last Glacial 319

Maximum. Nat. Geosc. 5, 74-79 (2012). 320

32 Beer, C. et al. Terrestrial Gross Carbon Dioxide Uptake: Global Distribution and Covariation 321

with Climate. Science 329, 834-838 (2010). 322

33 McMahon, S. M., Geoffrey G Parker, and Dawn R Miller. 2010. . Evidence for a recent 323

increase in forest growth. Proc. Nat. Acad. Sci 107, 3611–3615 (2010). 324

34 van der Sleen, P. et al. No growth stimulation of tropical trees by 150 Years of CO2

325

fertilization but water-use efficiency increased. Nat. Geosc. 8, 24–28 (2015). 326

35 Pugh, T. A. M., Muller, C., Arneth, A., Haverd, V. & Smith, B. Key knowledge and data gaps 327

in modelling the influence of CO2 concentration on the terrestrial carbon sink. J. Plant Phys. 328

203, 3-15 (2016). 329

36 Raymond, P. A. et al. Global carbon dioxide emissions from inland waters. Nature 503, 355-330

359 (2013). 331

15

37 Regnier, P. et al. Anthropogenic perturbation of the carbon fluxes from land to ocean. Nat. 332

Geosc. 6, 597-607, doi:10.1038/ngeo1830 (2013). 333

38 Prestele, R. et al. Hotspots of uncertainty in land use and land cover change projections: a 334

global scale model comparison. Glob. Change Biol., gcb.13337 (2016). 335

39 Bais, A. L. S., Lauk, C., Kastner, T. & Erb, K. Global patterns and trends of wood harvest and 336

use between 1990 and 2010. Ecol. Econ. 119, 326-337 (2015). 337

40 McGrath, M. J. et al. Reconstructing European forest management from 1600 to 2010. 338

Biogeosciences 12, 4291-4316 (2015). 339

41 Richter, D. D. & Houghton, R. A. Gross CO2 fluxes from land-use change: implications for 340

reducing global emissions and increasing sinks. Carb. Manag. 2, 41-47 (2011). 341

42 Bayer, A. D., Lindeskog, M., Pugh, T. A. M., Fuchs, R. & Arneth, A. Uncertainties in the land 342

use flux resulting from land use change reconstructions and gross land transitions. Earth Syst. 343

Dyn. Discuss. (2016). 344

43 Korner, C. Slow in, rapid out - Carbon flux studies and Kyoto targets. Science 300, 1242-1243 345

(2003). 346

44 Krause, A., Pugh, T. A. M., Bayer, A. D., Lindeskog, M. & Arneth, A. Impacts of land-use 347

history on the recovery of ecosystems after agricultural abandonment. Earth Syst. Dyn. 7, 745-348

766 (2016). 349

45 Zaehle, S., Jones, C. D., Houlton, B., Lamarque, J.-F. & Robertson, E. Nitrogen Availability 350

Reduces CMIP5 Projections of Twenty-First-Century Land Carbon Uptake. J. Clim. 28, 2494-351

2511 (2015). 352

46 Seidl, R., Schelhaas, M. J., Rammer, W. & Verkerk, P. J. Increasing forest disturbances in 353

Europe and their impact on carbon storage. Nat. Clim. Change 4, 806-810 (2014). 354

16

47 Hantson, S. et al. The status and challenge of global fire modelling. Biogeosciences 13, 3359-355

3375 (2016). 356

48 Running, S. W. Ecosystem disturbance, carbon, and climate. Science 321, 652-653 (2008). 357

49 Galik, C. S., Murray, B. C., Mitchell, S. & Cottle, P. Alternative approaches for addressing 358

non-permanence in carbon projects: an application to afforestation and reforestation under the 359

Clean Development Mechanism. Mitigation and Adaptation Strategies for Global Change 21, 360

101-118 (2016). 361

50 Friess, D. A., Phelps, J., Garmendia, E. & Gomez-Baggethun, E. Payments for Ecosystem 362

Services (PES) in the face of external biophysical stressors. Glob. Env. Change 30, 31-42 363 (2015). 364 365 Corresponding Author 366

Correspondence and request for materials should be addressed to Almut Arneth, 367 Almut.arneth@kit.edu 368 369 Acknowledgements 370

AA, ADB and TAMP acknowledge support from EU FP7 grants LUC4C (grant no. 603542) and 371

OPERAS (grant no.308393), and the Helmholtz Association in its ATMO programme and its impulse 372

and networking fund. MF, WL, CY and SS were also funded by LUC4C. JP and JEMSN were 373

supported by the German Research Foundation's Emmy Noether Program (PO 1751/1-1). EK was 374

supported by the ERTDF (S-10) from the Ministry of the Environment, Japan. ER was funded by 375

LUC4C and by the Joint UK DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). 376

SZ has received funding from the European Research Council (ERC) under the European Union‘s 377

Horizon 2020 research and innovation programme (grant agreement no. 647204; QUINCY). BDS is 378

17

supported by the Swiss National Science Foundation and FP7 funding through project EMBRACE 379

(282672). PC received support from the ERC SyG project IMBALANCE-P ‘Effects of phosphorus 380

limitations on Life, Earth system and Society’ Grant agreement no.: 610028.’ 381

382

Author contributions 383

AA, SS, JP, BS conceived the study. BP, LC, AB, MF, EK, JEMN, ADB, ML, TAMP, ER, TG, NV, 384

CY, SZ made changes to model code and provided simulation results. AA and SS analysed results. 385

BS, PC, WL provided Fig. 3. AA wrote the first draft, all authors commented on the draft and 386

discussion of results. 387

Additional information 388

Supplementary information is available in the online version of the paper. Reprints and permissions 389

information is available online at www.nature.com/reprints. Correspondence and requests for 390

materials should be addressed to A.A. 391

392

Box 1: Calculations of global terrestrial carbon uptake and removal 393

The net atmosphere-to-land carbon flux (FL) is generally inferred as the difference between other

394

terms of the global carbon cycle perturbation: 395

_
_
_  

 (1)

396

where
 _{are fossil fuel and cement emissions, FO is the atmosphere-ocean carbon exchange} 397

(currently an uptake) and  

 is the atmospheric growth rate of CO2 (equation 1).
 and  

 are 398

well known, and the estimate of the decadal global ocean carbon sink is bounded by a range of 399

observations1 such that the net land carbon flux is relatively well constrained. By contrast, there is 400

18

much less confidence in separating FL intoa carbon flux from anthropogenic land use and land cover

401

change (FLULCC), and a residual carbon flux to the land (FRL; (equation 2)) which is typically calculated

402

as the difference from the other carbon-cycle components: 403

_
_
_{ (2)}

404

FLULCC and FLR are both made up of source and sink fluxes. Uncertainties in FLULCC and FRL are around

405

35% - 40% over the period 1870-2014 (when expressed as % of the cumulative mean absolute values), 406

compared to 13% for the cumulative ocean sink and 5% for fossil fuel burning and cement emissions1. 407

FLULCC has been modelled by the bookkeeping method (combining data-driven representative carbon

408

stocks trajectories and/or, for the satellite period, remote-sensing information on carbon density for 409

different biomes, with estimates of land-cover change), or by dynamic global vegetation models 410

(DGVMs; calculating carbon density of ecosystems with process-based algorithms; see Methods). 411

DGVMs can also be used to calculate explicitly the magnitude and spatial distribution of FRL (refs 1,

412

3) instead of deducing its global value as a difference between FL andFLULCC as done in global budget

413

analyses. The bookkeeping approach has the advantage that carbon densities and carbon response 414

functions that describe the temporal evolution and fate of carbon after a LULCC disturbance can be 415

based directly on observational evidence6,23, but has to assume that local observations can be 416

extrapolated to regions/countries or biomes, thus partly ignoring spatial edaphic and climatic gradients 417

of carbon stocks. The DGVM-based simulations have the advantage to account for environmental 418

effects on carbon stocks through time, and account for spatial heterogeneity, but are poorly 419

constrained by data. DGVMs and bookkeeping models have similarly large degree of uncertainties1. 420

Figure captions

421

Figure 1: Difference in LULCC emission flux (∆FLULCC) due to individual processes. a, Wood harvest.

422

b, Shifting cultivation. c, Harvest (using the grass functional type). d, Full crop representation. 423

Coloured lines represent different models, grey symbols and hairlines are average ± one standard 424

deviation. 425

19

Figure 2: Response ratio of cumulative FLULCC,1 and FLULCC,0. See also Supplementary Table 1 and

426

methods for individual processes and models. 427

Figure 3: Comparison of net (a) and gross (b) forest / natural land change (million km2) between

428

different LULCC data sets. Changes in LUH1 data 7 represents the change of natural land because

429

there is no separate forest type in LUH1 while change in the other data sets indicates the forest change.

20 Methods 431

General simulation setup 432

Carbon fluxes from land-use change are derived as the difference between a simulation with 433

historically varying observed climate, atmospheric CO2 concentration and land-cover change (S3) and

434

one in which land-cover change was held constant (S2)1,2. Land-cover changes were taken from 435

HYDE51 or LUH17. In S2, land-cover distribution was fixed. Gridded historical estimates of gross-436

transitions (shifting cultivation (SC) in the tropics) and wood harvesting (WH) were taken from ref. 7. 437

Spin-up used repeated climate from the first decades of the twentieth century, and constant CO2

438

concentration and land-cover distribution (for details, see the individual description below). Upon 439

achieving steady-state, land-cover distribution and CO2 concentration were allowed to evolve

440

transiently, whereas transient climate evolution began at 1901. Atmospheric CO2 concentration was

441

taken from ice core data until roughly mid-twentieth century, when atmospheric measurements 442

became available1. A “baseline” carbon flux related to land-use change (FLULCC,0; see Supplementary

443

Table 1) is defined as excluding gross transitions and wood harvest, and using the grass plant 444

functional type to represent crop areas. Data in this Perspective were from previously published work, 445

supplemented by additional, new simulations. In cases where more than one of the processes that are 446

under investigation here were assessed by one model, several S3 experiments were provided. 447

Although spin-up and model configurations differed between models, for S2 and S3 simulations of 448

any one individual model the setup was the same, which allows to identify the effect of adding the 449

individual processes. We describe briefly the relevant aspects of models and simulational protocol, in 450

particular where they differ from their previously published versions. 451

JULES model. Here, to implement crop harvest, four additional PFTs were added: C3 crops, C4 452

crops, C3 pasture and C4 pasture, with identical parameter sets as the C3 and C4 grass PFTs. Lotka-453

Volterra equations52 are used three times to calculate the vegetation distribution in natural areas, crop 454

and pasture areas, with the calculations in each area being independent of the others. Crop-harvest is 455

21

represented by diverting 30% of crop litter to the fast product pool instead of to the soil; the fast 456

product pool has a rapid decay timescale of 1 year. Pasture is not harvested. 457

The model is forced by crop and pasture area from the Hyde 3.2 dataset1 and by CRU-NCEP 458

climate1,2, both at 1.875x1.25 degrees, using an hourly time-step, and updating vegetation distribution 459

every ten days. 1080 years of spin-up were run by fixing crop and pasture areas at 1860 levels and by 460

repeating 1901-1920 climate and CO2 concentrations.

461

JSBACH model. The JSBACH version used here is similar to the version in ref.1. S3 experiments 462

include gross land-use transitions and WH21. FLULCCc,0 in Supplementary Table 2 were calculated by

463

subtracting the individual contributions of these processes. Net transitions are derived from the gross 464

transition implementation, but by minimizing land conversions21. WH7 is taken not only from forest 465

PFTs but also shrubs and natural grasslands are harvested. Upon harvest, 20% of the carbon is 466

immediately released to the atmosphere; the rest is transferred into the litter and subject to soil 467

dynamics. JSBACH simulations were conducted at 1.9°x1.9° forced with remapped 1° LUH1 data 468

from 1860-2014 and daily climate calculated from the 6-hourly 0.5° CRU-NCEP product1 for the 469

years 1901-2014. The initial state in 1860 is based on a spin-up with 1860 CO2 concentrations (286.42

470

ppm), cycling (detrended) 1901-1921 climate and constant 1860 LUH1 WH amounts. From 1860 471

annual CO2 forcing was used, and after 1901climate was taken from CRU-NCEP. In the no-harvest

472

simulation the 1860 WH amounts were applied throughout the whole simulated period. 473

LPJ-GUESS model. SC: For implementing SC, recommendations followed those by ref.7, with 474

rotation periods of 15 years. Simulations used the coupled carbon-nitrogen version of the model16,53 475

spin-up used constant 1701 land-cover and CO2 concentration, and 1901-1930 recycled climate. Upon

476

steady-state land-cover and CO2 were allowed to change from 1701, and climate from 1901 onwards 42

. 477

When land is cleared, 76% of woody biomass and 71% of leaf biomass is removed and oxidised 478

within one year, with a further 21% of woody biomass assigned to a product pool with 25 year 479

turnover time42. Upon abandonment a secondary forest stand is created and recolonization of natural 480

22

vegetation takes place from a state of bare soil. With forest rotation, young stands (above a minimum 481

age of 15 years) are preferentially converted. 482

GH/MC: Simulations are taken from ref. 16, using the carbon-only version of the model. 68% of 483

deforested woody biomass and 75% of leaf biomass is oxidised within one year, with a further 30% of 484

woody biomass going to the product pool. In the GH case, 50% of the above-ground biomass are 485

annually removed from the ecosystem. In MC, 90% of the harvestable organs and an additional 75% 486

of above-ground crop residues are removed each year. Simulations ran from 1850 to 2012, with 1850 487

land-cover and CO2 concentrations, and recycled climate (1901-1930) being used for spin-up. All

LPJ-488

GUESS simulations used CRU TS 3.23 climate54. 489

LPJ model. Compared to previous versions, the model now uses the World Harmonization Soils 490

Database v.1.2 for soil texture and Cosby equations55 to estimate soil water holding capacity. Further 491

developments allow for gross land-use transitions and WH to be prescribed. Changes include: (1) the 492

primary grid-cell fraction only decreases in size; (2) secondary grid-cell fractions can decrease or 493

increase in size by combining with other secondary forest fractions, recently abandoned land, or 494

fractions with recent WH; (3) deforestation that results in an immediate flux to the atmosphere equal 495

to 100% of heartwood biomass and 50% of sapwood biomass; root biomass enters belowground litter 496

pools, while 100% leaf and 50% of sapwood biomass becomes part of aboveground litter. 497

WH demand7 on primary or secondary lands was met by the biomass in tree sapwood and heartwood 498

only. Only whole trees were harvested (that is, tree-density was reduced); wood from deforestation 499

was not included to meet WH demand. 100% of leaf biomass and 40% of the sapwood and heartwood 500

enters the aboveground litter, and 100% of root biomass enters the belowground litter pools; 60% of 501

sapwood and heartwood are assumed to go into a product pool. Of these, 55% go to the 1-year product 502

pool (emitted in the same year), 35% go to the 10-year product pool (emitted at rate 10% per year) and 503

10% go to the 100-year product pool (emitted at rate 1% per year). These delayed pool-emission 504

fluxes are part of the LULCC fluxes. After harvest, the harvested fraction is mixed with existing 505

secondary forest fraction, or a secondary fraction is created if none exists, while fully conserving 506

23

biomass. For simulations with SC, grid-cell fractions that underwent land-use change were not mixed 507

with existing managed lands or secondary fractions until all land-use transitions had occurred. 508

Simulations were performed using monthly CRU54 (TS v.3.23) climate at 0.5o degrees, and finished in 509

year 2013. Spin-up was done using recycled 1901-20 climate, and using 1860 land-cover and CO2.

510

Upon steady-state, land cover and CO2 varied after 1860 and climate varied after 1900.

511

LPJmL model. The LPJmL version used was as described in refs 56-58. In the baseline scenario all 512

crops were simulated as a mixture of C3 and C4 managed grasslands, 50% of the aboveground 513

biomass is transferred to the harvest compartment and assumed to be respired in the same year. 514

Climate data was 1901-2014 CRU TS v.3.23 monthly datasets and land-use patterns from the HYDE 515

3.2 dataset. Simulations were performed at 0.5o spatial resolution. Model spin-up used recycled 516

climate data from 1901-1920, and with land use patterns and CO2 concentrations fixed to the 1860

517

value. Simulations from 1861-2014 were done with varying annual CO2 concentration values, and

518

varying land use patterns according to the HYDE dataset, and with transient climate from 1901 until 519

2014. 520

LP3X model. Land-use change, including SC and WH, is implemented as described in ref. 20,

521

using the full land-use transition and wood harvesting data provided7. Wood (heartwood and

522

sapwood) removed by harvesting and land conversion is diverted to products pools with

523

turnover rates of 2 years (37.5%) and 20 years (37.5%). The rest, including slash from roots

524

and leaves is respired within the same year.

525

Simulation results shown here are based on employing the GCP 2015 protocol and input

526

data1. LPX includes interactive C and N cycling with N deposition and N fertiliser inputs59_.

527

Simulations with SC and WH were spun up to equilibrium under land-use transitions and WH

528

of year 150020_{. Varying land-use transitions and WH was included from 1500 onwards, with}

529

CO2 and N deposition of year 1860 and recycled climate from CRU TS v.3.23, years 1901-530

1931. All simulations are done on a 1 x 1 degree spatial resolution and make use of monthly

24

climate input. Original GCP standard input files were aggregated to 1 x 1 degrees conserving

532

area-weighted means (climate input) or absolute area of cropland and pasture (land use input).

533

OCN model. The OCN version used here is applied as in the framework of the annual carbon budget1. 534

OCN includes interactive C and N cycling with N deposition and N fertiliser inputs60. Wood harvest 535

was implemented by first satisfying the prescribed wood extraction rate from wood production due to 536

land-use change, and then removing additional biomass proportionally from forested tiles. Wood 537

(heartwood and sapwood) removed by harvesting and land conversion is diverted to products pools 538

with turnover rates of 1 years (59.7%), 10 years (40.2% for tropical, and 29.9% for extratropical trees) 539

and 100 years (10.4 % for extratropical trees)61. The remainder enters the litter pools. In case OCN’s 540

forest growth rate did not suffice to meet the prescribed wood extraction rate, harvesting was limited 541

to 5% of the total stand biomass and assumed to stop if the stand biomass density fell below 1 kg C m -542

2

. These limits were set to account for offsets in annual wood production between OCN’s predicted 543

biomass growth and the assumptions in the Hurtt et al. database7. These limits may lead to lower than 544

prescribed WH rates in low productive areas. An additional run was performed with keeping WH 545

constant at 1860s level. 546

Simulations with WH were spun up to equilibrium using harvesting of the year 18601. Varying land-547

use transitions or WH was included from 1860 onwards, with CO2 and N deposition of year 1860 and

548

recycled climate from CRU-NCEP, years 1901-1931. All simulations are done on a 1 x 1 degree 549

spatial resolution and make use of daily climate input, which is disaggregated to half-hourly values by 550

means of a weather generator62. Original GCP standard input files were aggregated to 1 x 1 degrees 551

conserving area-weighted means (climate input) or absolute area of cropland and pasture (land use 552

input). 553

554

ORCHIDEE model. WH: Developments to the version included in ref.1 include annual WH, the total 555

wood harvested of a grid cell is removed from above-ground biomass of the different forest PFTs 556

proportional (i) to its fraction in the gridcell and (ii) also to its relative biomass among forest PFTs. 557

25

This results in harvesting more wood in biomass-rich forests. In cases of inconsistencies between the 558

Orchidee and Hurtt forest fraction, and to avoid forest being degraded from excessive harvest we 559

assume that no more than 20% of the total forest biomass of a gridcell can be harvested in one year. 560

Hence the biomass actually harvested each year can be slightly lower than prescribed7. The harvested 561

biomass enters three pools of 1, 10 and 100 residence years respectively (and is part of FLULCC). Model

562

runs were done at 0.5°x0.5° resolution. Spin-up used recycled climate of 1901-1910. CO2

563

concentration, land-cover and WH of the year 1860. The model was run until the change in mean total 564

carbon of 98% of grid-points over a ten-year spin-up period was < 0.05%. 565

SC: Land cover transition matrices are upscaled from 0.5° LUH1 data7 so no transition information is 566

lost in the low-resolution run. The minimum bi-directional fluxes between two land cover types in 567

LUH1 were treated as shifting cultivation. The model was forced with CRU-NCEP forcing (v5.3.2), 568

re-gridded to 5° resolution from the original 0.5° resolution. Spin-up simulation used recycled climate 569

data for 1901-1910 with atmospheric CO2 held at 1750 level, and land cover fixed at 1500. Transient

570

runs started from 1501 until 2014, with CO2 varying from 1750 and climate varying from 1901. In the

571

transient run for the control simulation, land cover is held constant at 1500; for the SC run, land cover 572

varies by applying annual land use transition matrices of SC. All runs have been performed with 573

outputs on annual temporal resolution but forcing data is 6-hourly. 574

OSCAR model. A complete description of OSCAR v2.2 is provided by ref. 63. OSCAR is not a 575

DGVM, but a compact Earth system model calibrated on complex models. Here, it is used in an 576

offline setup in which the terrestrial carbon-cycle module is driven by exogenous changes in 577

atmospheric CO2 (IPCC AR5 WG1 Annex 2), climate (CRU TS v.3.23), and land-use and land cover

578

(HYDE v.3.2). 579

The global terrestrial biosphere is disaggregated into nine regions (detailed by ref. 64) and subdivided 580

into five biomes (bare soil, forest, shrubland+grassland, cropland, pasture). The carbon-cycle in each 581

of these 45 subparts is represented by a three-box model whose parameters are calibrated on DGVMs. 582

The preindustrial equilibrium (carbon densities and fluxes) is calibrated on TRENDY models2. The 583

26

transient response of NPP, heterotrophic respiration and wildfires to CO2 and/or climate is calibrated

584

on CMIP5 models65. The impact of land-use and land-cover change on the terrestrial carbon-cycle is 585

modelled using a bookkeeping approach. Coefficients used to allocate biomass after use or land-586

cover change are based on ref. 66. 587

Since OSCAR v2.2 is meant to be used in a probabilistic setup we made an ensemble of 2400 588

simulations in which the parameters (for example, preindustrial equilibrium, transient responses, 589

allocation coefficients) are drawn randomly from the pool of available parameterizations. See ref. 63 590

for more details. The resulting OSCAR values discussed and shown in the main text are the median of 591

this ensemble. 592

VISIT model. Implementation of climate, land-use change (gross transitions, SC) and WH has not 593

changed from1. Land-use, land-use change, and WH data for 1860-2014 were from LUH17. For WH, 594

the amount of harvested biomass prescribed in ref. 7 were transferred from simulated stem biomass to 595

1-year product pool (emitted in entirety in same year of wood harvest), 10-year product pool, and 100-596

year product pool in a same manner as in the cleared biomass with land-use change described in ref. 597

67. The non-harvested part of biomass remains in the ecosystem. The fluxes from WH pools are 598

included in the NBP calculations. 599

Climate data was 1901-2014 monthly CRU TS v.3.23 and all simulations were conducted with 0.5o 600

spatial resolution. The model spin-up was performed recycling climate data from 1901-1920, and with 601

land use patterns and CO2 concentrations fixed to the 1860 value. Simulations from 1860-2014 were

602

done with varying annual CO2 concentration values, varying land use patterns according to LUH1,

603

recycling the climate from 1901-1920 in the period 1860-1900, and with transient climate from 1901 604

until 2014. 605

606

Data in Fig. 3. Data for net forest change from the Food and Agriculture Organization (FAO)68 is

607

calculated as the difference of forest area between 2000 and 2010 in each region. The same data were

27

also used in the Houghton et al. bookkeeping model6. The net forest change from Hansen et al.69 is

609

based on satellite observations, andis their difference between gross forest gain and gross forest loss

610

during 2000-2012. Because the LUH1 dataset7 only has one type of natural vegetation, and does not

611

separate natural forest from natural grassland, the change in Fig. 3 represents the total change of

612

natural land. In Fig. 3b, for LUH1 the gross loss includes transitions from primary/secondary 613

vegetation to cropland / pasture, while the gross gain is the sum of transitions from cropland and

614

pasture to secondary land. With grasslands and forests treated as separate land-cover types in LUH2

615

(http://luh.umd.edu/), the change includes transitions from primary / secondary forest to cropland /

616

pasture (gross loss) and transitions from cropland / pasture to secondary forest (gross gain). The net

617

change for LUH1 or LUH2 is the difference between gross loss and gross gain. To be consistent with

618

ref. 69, the period calculated for LUH1 and LUH2 is also from 2000 to 2012. The products shown in 619

Figure 3 use definitions of forest loss and gain, and interpretation of differences between products 620

should therefore take these into consideration. 621

622

Data and code availability. The data that support the findings of this study are available upon 623

request, for access please contact almut.arneth@kit.edu and s.a.sitch@exeter.ac.uk. We are unable to 624

make the computer code of each of the models associated with this paper freely available because in 625

many cases the code is still under development. However, individual groups are open to share code 626

upon request, in case of interest please contact the co-authors for specific models. Access for LUH1 & 627

LUH2 is under http://luh.umd.edu/data.shtml; the HYDE data are accessible via

628

http://themasites.pbl.nl/tridion/en/themasites/hyde/download/index-2.html

629

28

References 631

51 Klein Goldewijk, L., Beusen, A., van Drecht, G. & de Vos, M. The HYDE 3.1 spatially 632

explicit database of human-induced global land use change over the past 12,000 years. Globl 633

Ecol. Biogeogr. 20, 73–86 (2011). 634

52 Clark, D. B. et al. The Joint UK Land Environment Simulator (JULES), model description – 635

Part 2: Carbon fluxes and vegetation dynamics. Geosci. Model Dev. 4, 701-722 (2011). 636

53 Smith, B. et al. Implications of incorporating N cycling and N limitations on primary 637

production in an individual-based dynamic vegetation model. Biogeosciences 11, 2027-2054 638

(2014). 639

54 Jones, P. & Harris, I. University of East Anglia Climatic Research Unit, CRU TS3. 21: 640

Climatic Research Unit (CRU) Time-Series (TS) Version 3.21 of High Resolution Gridded 641

Data of Month-by-month Variation in Climate (Jan. 1901—Dec. 2012). NCAS British 642

Atmospheric Data Centre (2013). 643

55 Cosby, B. J., Hornberger, G. M., Clapp, R. B. & Ginn, T. R. A STATISTICAL 644

EXPLORATION OF THE RELATIONSHIPS OF SOIL-MOISTURE CHARACTERISTICS 645

TO THE PHYSICAL-PROPERTIES OF SOILS. Water Resources Res. 20, 682-690 (1984). 646

56 Bondeau, A. et al. Modelling the role of agriculture for the 20th century global terrestrial 647

carbon balance. Glob. Change Biol. 13, 679-706 (2007). 648

57 Fader, M., von Bloh, W., Shi, S., Bondeau, A. & Cramer, W. Modelling Mediterranean agro-649

ecosystems by including agricultural trees in the LPJmL model. Geosc. Model Dev. 8, 3545-650

3561 (2015). 651

58 Waha, K., van Bussel, L. G. J., Müller, C. & Bondeau, A. Climate-driven simulation of global 652

crop sowing dates. Glob. Ecol. Biogeogr. 12, 247–259 (2012). 653

29

59 Stocker, B. D. et al. Multiple greenhouse-gas feedbacks from the land biosphere under future 654

climate change scenarios. Nat. Clim. Change 3, 666-672 (2013). 655

60 Zaehle, S., Ciais, P., Friend, A. D. & Prieur, V. Carbon benefits of anthropogenic reactive 656

nitrogen offset by nitrous oxide emissions. Nat. Geosc. 4, 601-605 (2011). 657

61 McGuire, A. D. et al. Carbon balance of the terrestrial biosphere in the twentieth century: 658

Analysis of CO2, climate and land use effects with four process-based ecosystem models.

659

Glob. Biogeochem. Cycles 15, 183-206 (2001). 660

62 Krinner, G., Ciais, P., Viovy, N. & Friedlingstein, P. A simple parameterization of nitrogen 661

limitation on primary productivity for global vegetation models. Biogeosciences Discussions 662

2, 1243-1282 (2005). 663

63 Gasser, T. et al. The compact Earth system model OSCAR v2.2: description and first results. 664

Geosc. Model Dev. submitted (2016). 665

64 Houghton, R. A. & Hackler, J. L. Carbon flux to the atmosphere from land-use changes: 1850 666

to 1990. (Carbon Dioxide Information Analysis Center, Oak Ridge, Tennessee, 2001). 667

65 Arora, V. K. et al. Carbon-Concentration and Carbon-Climate Feedbacks in CMIP5 Earth 668

System Models. J. Clim. 26, 5289-5314 (2013). 669

66 Mason, E. J., Yeh, S. & Skog, K. E. Timing of carbon emissions from global forest clearance. 670

Nature Clim. Change 2, 682-685 (2012). 671

67 Kato, E., Kinoshita, T., Ito, A., Kawamiya, M. & Yamagata, Y. Evaluation of spatially 672

explicit emission scenario of land-use change and biomass burning using a process-based 673

biogeochemical model. J. Land Use Sc. 8, 104-122 (2013). 674

68 Global Forest Resources Assessment 2010. (FAO, 2010). 675

30

69 Hansen, M. C. et al. High-Resolution Global Maps of 21st-Century Forest Cover Change. 676

Science 342, 850-853 (2013). 677

31 Figures 679 Figure 1 680 681

32 Figure 2 682 683 Figure 3 684 685